If you mostly use AI in a chat box and you are trying to decide which new tools are actually worth following, this is the part that matters. You see Old and new apps, via modern coding agents by Terry Tao, think it is another flashy build demo, and almost scroll past. That is the expensive mistake. If you only watch the surface hype, you burn time, budget, and attention in the wrong place. The real takeaway is not that AI can write code faster. It is this: coding agents, meaning AI tools that help carry a build forward step by step, are not replacing programmers. They are replacing the 'too hard, maybe later' decision.
Most people still judge AI coding with the wrong question: can it help me ship the same app faster? That is the loud version, but not the useful one. The better question is whether it lets you finish ideas you abandoned because the implementation burden made them feel not worth doing. A tool update is worth your attention only if it changes your next decision, not because it added more features.
That is why Tao's example lands. He wrote that he wanted to build a relativity visualization tool back in 1999, but gave up because the coding complexity was too high. In 2026, after a few hours of vibe coding with an AI agent, he finally got a version that matched the old vision [S001]. The point is not that the idea became smarter in 2026. The point is that a 27-year-old unfinished idea stopped being blocked by implementation friction.
The second useful detail is the boundary. In the making-of transcript for the spacetime diagram app, Tao was still doing the high-value human part: setting requirements, correcting the physics, and checking whether the output was actually right. Claude Code handled staged implementation and kept moving through testing [S004]. So this is not evidence that anyone can build complex software in a few hours. It is evidence that when you already know what 'done' should look like, a coding agent can remove a large part of the 'too hard' tax.
If you have an old app idea, a half-built tool, or a workflow you keep postponing because the build feels annoying, this is the filter to borrow. Do not ask, 'Can AI replace the builder?' Ask, 'What have I delayed because the build cost felt bigger than the idea?' Share this with the person who still has a good project sitting unfinished. That is who this story is actually for.